SpainA team of independent cybersecurity experts successfully accessed internal systems of OpenAI using Claude, an AI model developed by Anthropic, OpenAI’s main competitor. They hacked into the corporate ChatGPT account of an employee, allowing them to view internal files and propose changes to OpenAI’s repository of secret algorithms. This attack was conducted as part of OpenAI’s official bug bounty program, which encourages security researchers to find vulnerabilities before malicious actors do. While they did not alter the code of ChatGPT, they provided OpenAI with evidence of how such alterations could have been made. OpenAI rewarded the researchers $6,500 for their findings. The breach occurred in July but was recently revealed exclusively to The Wall Street Journal. The researchers, from the firm Hacktron AI, raised further concerns about the security practices of major AI laboratories. They used Claude in a way that violated Anthropic’s terms of service. OpenAI has acknowledged the researchers’ findings, while Anthropic has remained silent on the matter.
Bias read (Center): The article presents the event factually, without overtly favoring any side. It includes quotes from both OpenAI and mentions the involvement of Hacktron AI, providing a balanced perspective on the security breach and its implications. There is no clear ideological framing or biased language.
